How an AI learns · part 9 of 11 · Cross-validation, validation set and data leakage

What if you only have five examples?

Five students take turns in a test booth: one test student gives 100% or 0% by luck, while averaging every turn gives a steady 60% (k-fold cross-validation). Then a mock exam for tuning, a lucky 100% from tuning on the test, and two leaks: copied photos, and one patient's X-rays in both piles. Turn the number of folds and see how much the average moves.

What you turn: Number of turns (folds)

The reel